Tongue　image　classification　plays　a　very　important　role　in　the　Traditional　Chinese　Medicine　(TCM)　modernization.　However,　the　number　of　training　samples　labeled　by　the　authoritative　TCM　experts　is　small　since　it　is　a　hard　work　to　confirm　the　type　of　the　samples,　which　need　a　lot　of　time　and　human　labor.　Meantime　the　separable　boundary　obtained　by　these　labeled　samples　is　rough　and　imprecise.　Meantime　unlabeled　samples　are　abundant　and　easy　to　obtain.　Transductive　support　vector　machine　(TSVM)　is　a　method　to　reduce　human　labor　and　improve　accuracy　since　the　unlabeled　samples　can　be　joined　to　train　the　classifier　to　provide　much　more　classification　information　during　training.　The　experimental　results　show　that　the　TSVM　classifier　can　improve　the　right　classification　rate　so　it　is　a　promising　method　in　the　TCM　study.